troubles with overlay values

[Gianluca Roncalli]

Hi,

I’m trying to calculate the overlay between the seasonal home range of my birds by using the overlap function in ctmm package. I’ve a list of the different seasonal AKDE home ranges but the overlap matrix shows very weird values. For instance, even if the breeding and the wintering areas don’t overlap on the map, they have an overlay estimate which is different from 0 and, even more weird, the low and high levels of IC are improper, I guess.

I’ll attach the results of the overlay function, the map and also my ADKE r object. Maybe I’ve done something wrong.

Anybody can help me out?

Thank you

Best,

Gianluca

PS: Here's the script

load("bc181633_t_hr.Rdata")
bc181633<-hrange_list_sub2

overlap(bc181633, level=0.95)

 bc181633_t_hr.RData

--

Gianluca Roncalli

ISPRA (Istituto Superiore per la Protezione e la Ricerca Ambientale)

Area per l'Avifauna Migratrice (BIO-AVM)

Via Cà Fornacetta 9,  

40064 Ozzano dell'Emilia, BO 

Italy

 

ᐧ

[Christen Fleming]

Hi Gianluca,

The breeding-wintering overlap is estimated to be 0.3% (0.0%-94%), which initially seems strange for the upper CI, but if you look at the breeding model fit it's effective sample size is far below one: DOF[area] = 0.05. Meaning that this estimate has no certainty. Also, the numbers on the breeding season fit look very strange. The other fits are OUF/OUf, but breeding season is IID? I would check as to what is going on there.

Also, I really need to re-code the single grid calculation code for you to make this faster. I'll move that up the TODO list.

Best,

Chris

[Gianluca Roncalli]

Hi Chris,

Thanks for the prompt reply. 

Can you give us some more information about the meaning of DOF in the model summary table? I thought they indicate the degree of freedom of data used to calculate the area; don't they (DOF mean)? We have three column DOF mean, DOF area, and DOF speed. I've tried to find out some more detailed indications, but I couldn't find them. 

Thank you 

Best,

Gianluca

ᐧ

--
You received this message because you are subscribed to the Google Groups "ctmm R user group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ctmm-user+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ctmm-user/eef7fc79-2f98-405c-b44a-3e5cb4fac11ao%40googlegroups.com.

[Christen Fleming]

Hi Gianluca,

Yes, these are the degrees of freedom of the data (under the assumption of the autocorrelation model), which are also the effective sample sizes. Having DOF[area]<1 is very bad for area estimation. With this model being IID, in contrast to the others being OUF/OUf, I would check that this part of the analysis has come out correctly.

Best,

Chris

To unsubscribe from this group and stop receiving emails from it, send an email to ctmm-user+unsubscribe@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ctmm-user/eef7fc79-2f98-405c-b44a-3e5cb4fac11ao%40googlegroups.com.

[Christen Fleming]

The updated package on GitHub should take a lot less memory when calculating spread out AKDEs on the compatible grids. There's still room for improvement, but this should help a bit.

Best,

Chris

[Gianluca Roncalli]

Thanks. I actually had a lot of problems when selecting the models and the spread of the home ranges. So, I guess this is a very good improvement. 

Best,

Gianlu

--

You received this message because you are subscribed to the Google Groups "ctmm R user group" group.

To unsubscribe from this group and stop receiving emails from it, send an email to ctmm-user+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ctmm-user/88578f93-967b-460e-8240-97c585d82ad1o%40googlegroups.com.

[roncalli...@gmail.com]

Hi,

I appreciated your explanation regarding the HR calculation and, particularly with respect to the importance of pay attention to DOF values. Nevertheless, I still have some problems (very weird overlap values) for 4 of my animals. The low and up CIs have an extremely large range (for example, in one case, I obtained these values: 0, 0.52, 1), even if all the parameters selected by the best AIC model look good.

I would attach the calculated AKDE as Rdata object in order to understand if I did something wrong and to have some advice. Do you need more data?

Thank you

Best

Gianlu 

[Christen Fleming]

Hi Gianluca,

I can take a look if you need, but you should expect relatively wide confidence intervals on the overlap estimates if either of the constituent distributions also have relatively wide confidence intervals (small DOF[mean] or small DOF[area]).

Best,

Chris